1 From de3de54b1610dfac21bf571fa03ccee3d94955df Mon Sep 17 00:00:00 2001
2 From: Steven Rostedt <rostedt@goodmis.org>
3 Date: Thu, 5 Dec 2013 09:16:52 -0500
4 Subject: [PATCH 220/365] cpu hotplug: Document why PREEMPT_RT uses a spinlock
8 cpu: Make hotplug.lock a "sleeping" spinlock on RT
10 Tasks can block on hotplug.lock in pin_current_cpu(), but their
11 state might be != RUNNING. So the mutex wakeup will set the state
12 unconditionally to RUNNING. That might cause spurious unexpected
13 wakeups. We could provide a state preserving mutex_lock() function,
14 but this is semantically backwards. So instead we convert the
15 hotplug.lock() to a spinlock for RT, which has the state preserving
18 Fixed a bug where the hotplug lock on PREEMPT_RT can be called after a
19 task set its state to TASK_UNINTERRUPTIBLE and before it called
20 schedule. If the hotplug_lock used a mutex, and there was contention,
21 the current task's state would be turned to TASK_RUNNABLE and the
22 schedule call will not sleep. This caused unexpected results.
24 Although the patch had a description of the change, the code had no
25 comments about it. This causes confusion to those that review the code,
26 and as PREEMPT_RT is held in a quilt queue and not git, it's not as easy
27 to see why a change was made. Even if it was in git, the code should
28 still have a comment for something as subtle as this.
30 Document the rational for using a spinlock on PREEMPT_RT in the hotplug
33 Reported-by: Nicholas Mc Guire <der.herr@hofr.at>
34 Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
35 Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
37 kernel/cpu.c | 8 ++++++++
38 1 file changed, 8 insertions(+)
40 diff --git a/kernel/cpu.c b/kernel/cpu.c
41 index d3e2928..4522046 100644
44 @@ -112,6 +112,14 @@ struct hotplug_pcp {
46 struct completion synced;
47 #ifdef CONFIG_PREEMPT_RT_FULL
49 + * Note, on PREEMPT_RT, the hotplug lock must save the state of
50 + * the task, otherwise the mutex will cause the task to fail
51 + * to sleep when required. (Because it's called from migrate_disable())
53 + * The spinlock_t on PREEMPT_RT is a mutex that saves the task's